[ES\QL] Text embedding function constant folding

[afoucret]

This PR introduces constant folding optimization for the TEXT_EMBEDDING function in ESQL.

Key components:

• InferenceFunctionEvaluator: a new piece of infrastructure that uses an InferenceOperator to evaluate an inference function. Currently, only folding of constants is supported, but it could be extended to more use cases in the future.

• Added a rules mechanism to the LogicalPlanPreOptimizer and introduced a rule that folds inference functions (FoldInferenceFunctions)

• Adding CSV tests for the TEXT_EMBEDDING function, including usage with other vector functions such as KNN)

Part of #131022

[elasticsearchmachine]

Pinging @elastic/es-search-relevance (Team:Search Relevance)

[github-actions]

ℹ️ Important: Docs version tagging

👋 Thanks for updating the docs! Just a friendly reminder that our docs are now cumulative. This means all 9.x versions are documented on the same page and published off of the main branch, instead of creating separate pages for each minor version.

We use applies_to tags to mark version-specific features and changes.

Expand for a quick overview

When to use applies_to tags:

✅ At the page level to indicate which products/deployments the content applies to (mandatory)
✅ When features change state (e.g. preview, ga) in a specific version
✅ When availability differs across deployments and environments

What NOT to do:

❌ Don't remove or replace information that applies to an older version
❌ Don't add new information that applies to a specific version without an applies_to tag
❌ Don't forget that applies_to tags can be used at the page, section, and inline level

🤔 Need help?

• Check out the cumulative docs guidelines
• Reach out in the #docs Slack channel

[afoucret]

ℹ️ This tests were not added before because they were an issue while the dense vector type was under construction.

[afoucret]

ℹ️ Doc update are caused by the dense vector type being out of snapshot.

[ioanatia]

can we get a test with multiple text_embedding calls with different query strings?

[afoucret]

I added a CSV test with FORK that should be covering your ask:

FROM semantic_text METADATA _score
| FORK (EVAL query_embedding = TEXT_EMBEDDING("be excellent to each other", "test_dense_inference") | WHERE KNN(semantic_text_dense_field, query_embedding))
       (EVAL query_embedding = TEXT_EMBEDDING("live long and prosper", "test_dense_inference") | WHERE KNN(semantic_text_dense_field, query_embedding))
| KEEP semantic_text_field, query_embedding, _score, _fork
| EVAL _score = ROUND(_score, 4)
| SORT _score DESC
| LIMIT 10
;

semantic_text_field:text                                              | query_embedding:dense_vector | _fork:keyword | _score:double
be excellent to each other                                            | [45.0, 55.0, 54.0]           | fork1         | 1.0
live long and prosper                                                 | [50.0, 57.0, 56.0]           | fork2         | 1.0
be excellent to each other                                            | [50.0, 57.0, 56.0]           | fork2         | 0.0295
live long and prosper                                                 | [45.0, 55.0, 54.0]           | fork1         | 0.0295
all we have to decide is what to do with the time that is given to us | [45.0, 55.0, 54.0]           | fork1         | 0.0214
all we have to decide is what to do with the time that is given to us | [50.0, 57.0, 56.0]           | fork2         | 0.0109

[afoucret]

plus it tests inference function in the context of fork

[carlosdelest]

LGTM 💯

[ioanatia]

I guess something like this would also be used for folding COMPLETION when it's used with a foldable prompt?

[afoucret]

You are right. There will be something similar for inference plans.

[ioanatia]

should these be skipped for release tests?

[afoucret]

I double-checked and it is not necessary.
The test is passing in release build since it does not use features that change in release build (function registry, writeable, ...)